@adhdev/daemon-core 0.9.82-rc.9 → 0.9.82-rc.90

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (67) hide show
  1. package/dist/cli-adapters/provider-cli-adapter.d.ts +2 -0
  2. package/dist/cli-adapters/provider-cli-parse.d.ts +1 -0
  3. package/dist/cli-adapters/provider-cli-shared.d.ts +2 -0
  4. package/dist/commands/router.d.ts +22 -0
  5. package/dist/config/mesh-config.d.ts +66 -1
  6. package/dist/index.d.ts +13 -6
  7. package/dist/index.js +5395 -1197
  8. package/dist/index.js.map +1 -1
  9. package/dist/index.mjs +5359 -1183
  10. package/dist/index.mjs.map +1 -1
  11. package/dist/installer.d.ts +1 -4
  12. package/dist/launch.d.ts +1 -1
  13. package/dist/logging/async-batch-writer.d.ts +10 -0
  14. package/dist/mesh/beads-db.d.ts +18 -0
  15. package/dist/mesh/mesh-active-work.d.ts +60 -0
  16. package/dist/mesh/mesh-events.d.ts +29 -5
  17. package/dist/mesh/mesh-fast-forward.d.ts +39 -0
  18. package/dist/mesh/mesh-host-ownership.d.ts +9 -0
  19. package/dist/mesh/mesh-ledger.d.ts +38 -1
  20. package/dist/mesh/mesh-work-queue.d.ts +27 -5
  21. package/dist/mesh/refine-config.d.ts +176 -0
  22. package/dist/providers/chat-message-normalization.d.ts +1 -0
  23. package/dist/providers/cli-provider-instance.d.ts +2 -1
  24. package/dist/repo-mesh-types.d.ts +46 -0
  25. package/dist/status/reporter.d.ts +2 -0
  26. package/package.json +3 -1
  27. package/src/boot/daemon-lifecycle.ts +1 -0
  28. package/src/cli-adapters/provider-cli-adapter.ts +91 -3
  29. package/src/cli-adapters/provider-cli-parse.d.ts +1 -0
  30. package/src/cli-adapters/provider-cli-parse.ts +4 -0
  31. package/src/cli-adapters/provider-cli-runtime.ts +3 -1
  32. package/src/cli-adapters/provider-cli-shared.d.ts +2 -0
  33. package/src/cli-adapters/provider-cli-shared.ts +20 -10
  34. package/src/commands/chat-commands.ts +454 -15
  35. package/src/commands/cli-manager.ts +126 -0
  36. package/src/commands/handler.ts +8 -1
  37. package/src/commands/mesh-coordinator.ts +13 -143
  38. package/src/commands/router.ts +2687 -435
  39. package/src/config/chat-history.ts +9 -7
  40. package/src/config/mesh-config.ts +245 -1
  41. package/src/daemon/dev-cli-debug.ts +10 -1
  42. package/src/detection/ide-detector.ts +26 -16
  43. package/src/index.ts +31 -5
  44. package/src/installer.d.ts +1 -1
  45. package/src/installer.ts +8 -6
  46. package/src/launch.d.ts +1 -1
  47. package/src/launch.ts +37 -28
  48. package/src/logging/async-batch-writer.ts +55 -0
  49. package/src/logging/logger.ts +2 -1
  50. package/src/mesh/beads-db.ts +176 -0
  51. package/src/mesh/coordinator-prompt.ts +30 -7
  52. package/src/mesh/mesh-active-work.ts +243 -0
  53. package/src/mesh/mesh-events.ts +400 -47
  54. package/src/mesh/mesh-fast-forward.ts +430 -0
  55. package/src/mesh/mesh-host-ownership.ts +73 -0
  56. package/src/mesh/mesh-ledger.ts +138 -1
  57. package/src/mesh/mesh-work-queue.ts +199 -137
  58. package/src/mesh/refine-config.ts +356 -0
  59. package/src/providers/chat-message-normalization.ts +3 -1
  60. package/src/providers/cli-provider-instance.ts +91 -13
  61. package/src/providers/ide-provider-instance.ts +17 -3
  62. package/src/providers/provider-loader.ts +10 -4
  63. package/src/providers/read-chat-contract.ts +1 -1
  64. package/src/providers/version-archive.ts +38 -20
  65. package/src/repo-mesh-types.ts +51 -0
  66. package/src/status/reporter.ts +15 -0
  67. package/src/system/host-memory.ts +29 -12
@@ -1,63 +1,179 @@
1
+ import { appendFileSync, existsSync, readFileSync, unlinkSync } from 'fs';
2
+ import { join } from 'path';
1
3
  import type { DaemonComponents } from '../boot/daemon-lifecycle.js';
2
4
  import { loadConfig } from '../config/config.js';
3
5
  import { getMesh, getMeshByRepo } from '../config/mesh-config.js';
4
6
  import { detectCLI } from '../detection/cli-detector.js';
5
7
  import { LOG } from '../logging/logger.js';
6
- import { appendLedgerEntry, buildTaskCompletionEvidence, getSessionRecoveryContext, isIntentionalCleanupStopEntry, readLedgerEntries } from './mesh-ledger.js';
8
+ import { appendLedgerEntry, buildTaskCompletionEvidence, getLedgerDir, getSessionRecoveryContext, isIntentionalCleanupStopEntry, readLedgerEntries } from './mesh-ledger.js';
7
9
  import type { MeshLedgerKind, SessionRecoveryContext } from './mesh-ledger.js';
8
10
  import { claimNextTask, updateSessionTaskStatus, enqueueTask, updateTaskStatus, getQueue, recordTaskAutoLaunch } from './mesh-work-queue.js';
9
11
 
10
12
  // ---------------------------------------------------------------------------
11
13
  // Remote Node Idle Session Tracking
12
14
  // ---------------------------------------------------------------------------
13
- // Tracks remote sessions that emitted 'agent:ready' so triggerMeshQueue
14
- // can assign tasks to them.
15
+ // Tracks remote sessions that emitted 'agent:ready' so triggerMeshQueue
16
+ // can assign tasks to them. Each entry carries an expiresAt timestamp;
17
+ // entries are swept on insertion to prevent unbounded growth.
15
18
  // ---------------------------------------------------------------------------
16
19
  interface RemoteIdleSession {
17
20
  nodeId: string;
18
21
  sessionId: string;
19
22
  providerType: string;
23
+ expiresAt: number;
20
24
  }
25
+ const REMOTE_IDLE_SESSION_TTL_MS = 5 * 60 * 1000; // 5 minutes
21
26
  const remoteIdleSessions = new Map<string, RemoteIdleSession>(); // key: `${nodeId}:${sessionId}`
22
27
 
28
+ function readWorkerResultMetadata(event: Record<string, unknown>): Record<string, unknown> | undefined {
29
+ return readRecord(event.workerResult) || readRecord(event.meshWorkerResult) || readRecord(event.structuredResult);
30
+ }
31
+
32
+ function sweepExpiredRemoteIdleSessions(): void {
33
+ const now = Date.now();
34
+ for (const [key, session] of remoteIdleSessions) {
35
+ if (session.expiresAt <= now) remoteIdleSessions.delete(key);
36
+ }
37
+ }
38
+
23
39
  // ---------------------------------------------------------------------------
24
- // MCP coordinator pending-event queue
40
+ // MCP coordinator pending-event queue — FILE-BASED PERSISTENCE
25
41
  // ---------------------------------------------------------------------------
26
42
  // When a mesh event fires but no CLI coordinator session is registered (e.g.
27
- // the coordinator is Claude Code running via MCP), we buffer the event here.
28
- // The MCP server drains this queue on every mesh_status / mesh_send_task poll.
43
+ // the coordinator is Claude Code running via MCP), we persist the event to a
44
+ // per-mesh JSONL file so it survives daemon restarts. The 50-entry hard cap
45
+ // is removed; the file is drained atomically on each get_pending_mesh_events
46
+ // call and limited to 100 KB to prevent runaway growth.
47
+ //
48
+ // File: <ledgerDir>/<meshId>.pending-events.jsonl
29
49
  // ---------------------------------------------------------------------------
30
50
 
31
51
  export interface PendingMeshCoordinatorEvent {
32
52
  event: string;
33
53
  meshId: string;
34
54
  nodeLabel: string;
55
+ nodeId?: string;
56
+ workspace?: string;
35
57
  metadataEvent: Record<string, unknown>;
58
+ coordinatorMessage?: string;
36
59
  queuedAt: number;
37
60
  }
38
61
 
39
- const MAX_PENDING_EVENTS = 50;
40
- const pendingMeshCoordinatorEvents: PendingMeshCoordinatorEvent[] = [];
62
+ const REFINE_TERMINAL_EVENTS = new Set(['refine:completed', 'refine:failed']);
63
+
64
+ function readRefineJobId(event: { metadataEvent?: Record<string, unknown> } | Record<string, unknown>): string {
65
+ const metadata = readRecord((event as any).metadataEvent) || event as Record<string, unknown>;
66
+ const result = readRecord(metadata.result);
67
+ const refineJob = readRecord(result?.refineJob);
68
+ return readNonEmptyString(metadata.jobId) || readNonEmptyString(refineJob?.jobId);
69
+ }
70
+
71
+ function buildRefineTerminalEventFingerprint(meshId: string, eventName: string, metadataEvent: Record<string, unknown>): string {
72
+ const jobId = readRefineJobId({ metadataEvent });
73
+ return jobId && REFINE_TERMINAL_EVENTS.has(eventName) ? `${meshId}::${eventName}::${jobId}` : '';
74
+ }
75
+
76
+ function hasPendingRefineTerminalEventDuplicate(event: PendingMeshCoordinatorEvent): boolean {
77
+ if (!REFINE_TERMINAL_EVENTS.has(event.event)) return false;
78
+ const jobId = readRefineJobId(event);
79
+ if (!jobId) return false;
80
+ return getPendingMeshCoordinatorEvents(event.meshId).some((pending) =>
81
+ pending.event === event.event && readRefineJobId(pending) === jobId,
82
+ );
83
+ }
84
+
85
+ function buildPendingEventFingerprint(event: PendingMeshCoordinatorEvent): string {
86
+ const metadata = readRecord(event.metadataEvent) || {};
87
+ const sessionId = resolveEventSessionId(metadata);
88
+ const providerSessionId = readNonEmptyString(metadata.providerSessionId);
89
+ const taskId = readNonEmptyString(metadata.taskId) || readNonEmptyString(readRecord(metadata.payload)?.taskId);
90
+ const jobId = readRefineJobId(event);
91
+ const timestamp = metadata.timestamp !== undefined && metadata.timestamp !== null ? String(metadata.timestamp) : '';
92
+ return [
93
+ event.meshId,
94
+ event.event,
95
+ event.nodeId || '',
96
+ sessionId || '',
97
+ providerSessionId || '',
98
+ taskId || '',
99
+ jobId || '',
100
+ timestamp || '',
101
+ ].join('::');
102
+ }
103
+
104
+ function hasPendingCoordinatorEventDuplicate(event: PendingMeshCoordinatorEvent): boolean {
105
+ const fingerprint = buildPendingEventFingerprint(event);
106
+ if (!fingerprint.trim()) return false;
107
+ return getPendingMeshCoordinatorEvents(event.meshId).some((pending) => buildPendingEventFingerprint(pending) === fingerprint);
108
+ }
109
+
110
+ function getPendingEventsPath(meshId: string): string {
111
+ const safe = meshId.replace(/[^a-zA-Z0-9_-]/g, '_');
112
+ return join(getLedgerDir(), `${safe}.pending-events.jsonl`);
113
+ }
114
+
115
+ export function queuePendingMeshCoordinatorEvent(event: PendingMeshCoordinatorEvent): boolean {
116
+ try {
117
+ if (hasPendingRefineTerminalEventDuplicate(event)) {
118
+ LOG.info('MeshEvents', `Suppressed duplicate pending ${event.event} for refine job ${readRefineJobId(event)}`);
119
+ return true;
120
+ }
121
+ if (hasPendingCoordinatorEventDuplicate(event)) {
122
+ LOG.info('MeshEvents', `Suppressed duplicate pending ${event.event} for mesh ${event.meshId}`);
123
+ return true;
124
+ }
125
+ appendFileSync(getPendingEventsPath(event.meshId), JSON.stringify(event) + '\n', 'utf-8');
126
+ return true;
127
+ } catch (e: any) {
128
+ LOG.warn('MeshEvents', `Failed to persist pending coordinator event: ${e?.message || e}`);
129
+ return false;
130
+ }
131
+ }
41
132
 
42
- /** Drain and return all pending coordinator events, clearing the queue. */
43
- export function drainPendingMeshCoordinatorEvents(): PendingMeshCoordinatorEvent[] {
44
- return pendingMeshCoordinatorEvents.splice(0);
133
+ /** Drain and return all pending coordinator events for meshId, removing them from disk. */
134
+ export function drainPendingMeshCoordinatorEvents(meshId?: string): PendingMeshCoordinatorEvent[] {
135
+ if (!meshId) return [];
136
+ const path = getPendingEventsPath(meshId);
137
+ if (!existsSync(path)) return [];
138
+ try {
139
+ const raw = readFileSync(path, 'utf-8');
140
+ try { unlinkSync(path); } catch { /* concurrent drain already removed it */ }
141
+ return raw.split('\n').filter(Boolean).flatMap(line => {
142
+ try { return [JSON.parse(line) as PendingMeshCoordinatorEvent]; } catch { return []; }
143
+ });
144
+ } catch { return []; }
45
145
  }
46
146
 
47
147
  /** Peek at pending coordinator events without draining (non-destructive). */
48
- export function getPendingMeshCoordinatorEvents(): readonly PendingMeshCoordinatorEvent[] {
49
- return pendingMeshCoordinatorEvents.slice();
148
+ export function getPendingMeshCoordinatorEvents(meshId?: string): readonly PendingMeshCoordinatorEvent[] {
149
+ if (!meshId) return [];
150
+ const path = getPendingEventsPath(meshId);
151
+ if (!existsSync(path)) return [];
152
+ try {
153
+ const raw = readFileSync(path, 'utf-8');
154
+ return raw.split('\n').filter(Boolean).flatMap(line => {
155
+ try { return [JSON.parse(line) as PendingMeshCoordinatorEvent]; } catch { return []; }
156
+ });
157
+ } catch { return []; }
50
158
  }
51
159
 
52
- /** Explicitly clear all pending coordinator events. */
53
- export function clearPendingMeshCoordinatorEvents(): void {
54
- pendingMeshCoordinatorEvents.splice(0);
160
+ /** Explicitly clear all pending coordinator events for a mesh. */
161
+ export function clearPendingMeshCoordinatorEvents(meshId?: string): void {
162
+ if (!meshId) return;
163
+ const path = getPendingEventsPath(meshId);
164
+ if (existsSync(path)) try { unlinkSync(path); } catch { /* already removed */ }
55
165
  }
56
166
 
57
167
  function readNonEmptyString(value: unknown): string {
58
168
  return typeof value === 'string' && value.trim() ? value.trim() : '';
59
169
  }
60
170
 
171
+ function readRecord(value: unknown): Record<string, unknown> | undefined {
172
+ return value && typeof value === 'object' && !Array.isArray(value)
173
+ ? value as Record<string, unknown>
174
+ : undefined;
175
+ }
176
+
61
177
  function resolveEventSessionId(event: Record<string, unknown>, fallback?: unknown): string {
62
178
  return readNonEmptyString(event.targetSessionId)
63
179
  || readNonEmptyString(event.sessionId)
@@ -72,6 +188,9 @@ const MESH_COORDINATOR_EVENTS = new Set([
72
188
  'agent:stopped',
73
189
  'agent:ready',
74
190
  'monitor:long_generating',
191
+ 'refine:accepted',
192
+ 'refine:completed',
193
+ 'refine:failed',
75
194
  ]);
76
195
 
77
196
  const EVENT_TO_LEDGER_KIND: Record<string, MeshLedgerKind> = {
@@ -86,10 +205,21 @@ function isMeshCoordinatorEvent(eventName: unknown): eventName is string {
86
205
  }
87
206
 
88
207
  function formatCompletionMetadata(event: Record<string, unknown>): string {
208
+ const completionDiagnostic = event.completionDiagnostic && typeof event.completionDiagnostic === 'object'
209
+ ? event.completionDiagnostic as Record<string, unknown>
210
+ : null;
211
+ const diagnosticReason = completionDiagnostic
212
+ ? readNonEmptyString(completionDiagnostic.blockReason) || 'present'
213
+ : '';
214
+ const finalAssistantPresent = typeof completionDiagnostic?.finalAssistantPresent === 'boolean'
215
+ ? String(completionDiagnostic.finalAssistantPresent)
216
+ : '';
89
217
  const parts = [
90
218
  readNonEmptyString(event.targetSessionId) ? `session_id=${readNonEmptyString(event.targetSessionId)}` : '',
91
219
  readNonEmptyString(event.providerType) ? `provider=${readNonEmptyString(event.providerType)}` : '',
92
220
  readNonEmptyString(event.providerSessionId) ? `provider_session_id=${readNonEmptyString(event.providerSessionId)}` : '',
221
+ diagnosticReason ? `completion_diagnostic=${diagnosticReason}` : '',
222
+ finalAssistantPresent ? `final_assistant=${finalAssistantPresent}` : '',
93
223
  ].filter(Boolean);
94
224
  return parts.length > 0 ? ` (${parts.join('; ')})` : '';
95
225
  }
@@ -140,6 +270,74 @@ function shouldSuppressIntentionalCleanupStop(args: {
140
270
  return hasRecentIntentionalCleanupStop(args.meshId, args.sessionId, args.nodeId);
141
271
  }
142
272
 
273
+ const RECENT_COMPLETION_FINGERPRINT_TTL_MS = 10 * 60 * 1000;
274
+ const recentCompletionFingerprints = new Map<string, number>();
275
+
276
+ function readEventTimestamp(value: unknown): number | null {
277
+ if (typeof value === 'number' && Number.isFinite(value)) return value;
278
+ if (typeof value === 'string' && value.trim()) {
279
+ const numeric = Number(value);
280
+ if (Number.isFinite(numeric)) return numeric;
281
+ const parsed = Date.parse(value);
282
+ if (Number.isFinite(parsed)) return parsed;
283
+ }
284
+ return null;
285
+ }
286
+
287
+ function buildMeshCompletionFingerprint(args: {
288
+ meshId: string;
289
+ event: string;
290
+ sessionId: string;
291
+ providerType?: string;
292
+ providerSessionId?: string;
293
+ timestamp?: number | null;
294
+ finalSummary?: string;
295
+ }): string {
296
+ const timestampPart = Number.isFinite(args.timestamp)
297
+ ? String(args.timestamp)
298
+ : readNonEmptyString(args.finalSummary).slice(0, 200);
299
+ return [
300
+ args.meshId,
301
+ args.event,
302
+ args.sessionId,
303
+ args.providerType || '',
304
+ args.providerSessionId || '',
305
+ timestampPart,
306
+ ].join('::');
307
+ }
308
+
309
+ function isDuplicateMeshCompletionEvent(args: {
310
+ meshId: string;
311
+ event: string;
312
+ sessionId: string;
313
+ providerType?: string;
314
+ providerSessionId?: string;
315
+ timestamp?: number | null;
316
+ finalSummary?: string;
317
+ }): boolean {
318
+ const fingerprint = buildMeshCompletionFingerprint(args);
319
+ if (!fingerprint) return false;
320
+ const now = Date.now();
321
+ for (const [key, seenAt] of recentCompletionFingerprints.entries()) {
322
+ if (now - seenAt > RECENT_COMPLETION_FINGERPRINT_TTL_MS) recentCompletionFingerprints.delete(key);
323
+ }
324
+ if (recentCompletionFingerprints.has(fingerprint)) return true;
325
+ recentCompletionFingerprints.set(fingerprint, now);
326
+ return false;
327
+ }
328
+
329
+ function isDuplicateRefineTerminalEvent(meshId: string, eventName: string, metadataEvent: Record<string, unknown>): boolean {
330
+ const fingerprint = buildRefineTerminalEventFingerprint(meshId, eventName, metadataEvent);
331
+ if (!fingerprint) return false;
332
+ const now = Date.now();
333
+ for (const [key, seenAt] of recentCompletionFingerprints.entries()) {
334
+ if (now - seenAt > RECENT_COMPLETION_FINGERPRINT_TTL_MS) recentCompletionFingerprints.delete(key);
335
+ }
336
+ if (recentCompletionFingerprints.has(fingerprint)) return true;
337
+ recentCompletionFingerprints.set(fingerprint, now);
338
+ return false;
339
+ }
340
+
143
341
 
144
342
  export function tryAssignQueueTask(
145
343
  components: DaemonComponents,
@@ -170,7 +368,16 @@ export function tryAssignQueueTask(
170
368
  message: task.message,
171
369
  }).catch((e: any) => {
172
370
  LOG.error('MeshQueue', `Failed to dispatch task via P2P to remote node ${nodeId}: ${e?.message}`);
173
- updateTaskStatus(meshId, task.id, 'failed');
371
+ // Revert to pending so the task can be retried rather than permanently failing
372
+ updateTaskStatus(meshId, task.id, 'pending');
373
+ try {
374
+ appendLedgerEntry(meshId, {
375
+ kind: 'dispatch_failed' as any,
376
+ nodeId,
377
+ sessionId,
378
+ payload: { taskId: task.id, error: e?.message, retryable: true },
379
+ });
380
+ } catch { /* ledger write is best-effort */ }
174
381
  });
175
382
  return true;
176
383
  }
@@ -565,6 +772,71 @@ function buildMeshSystemMessage(args: {
565
772
  if (args.event === 'monitor:long_generating') {
566
773
  return `[System] ${args.nodeLabel} has been generating for a long time${metadata}. Use mesh_read_chat once for a status check, but do not poll repeatedly.`;
567
774
  }
775
+ if (args.event === 'refine:accepted') {
776
+ const jobId = readRefineJobId({ metadataEvent: args.metadataEvent });
777
+ return `[System] Refinery accepted async job${jobId ? ` ${jobId}` : ''} for ${args.nodeLabel}. Completion/failure will be delivered as a terminal refine event; do not poll repeatedly.`;
778
+ }
779
+ if (args.event === 'refine:completed') {
780
+ const jobId = readRefineJobId({ metadataEvent: args.metadataEvent });
781
+ const result = readRecord(args.metadataEvent.result);
782
+ const validationSummary = readRecord(result?.validationSummary);
783
+ const patchEquivalence = readRecord(result?.patchEquivalence);
784
+ const finalConvergence = readRecord(result?.finalBranchConvergenceState);
785
+ const validationStatus = readNonEmptyString(validationSummary?.status);
786
+ const patchStatus = readNonEmptyString(patchEquivalence?.status)
787
+ || (patchEquivalence?.equivalent === true ? 'passed' : '');
788
+ const into = readNonEmptyString(result?.into);
789
+ const branch = readNonEmptyString(result?.branch);
790
+ const mergeStatus = result?.merged === true ? 'merged' : readNonEmptyString(finalConvergence?.status);
791
+ const convergenceStatus = readNonEmptyString(finalConvergence?.status);
792
+ const nextStep = readNonEmptyString(result?.nextStep)
793
+ || readNonEmptyString(finalConvergence?.nextStep)
794
+ || 'Continue from the updated mesh state.';
795
+ const details = [
796
+ jobId ? `job_id=${jobId}` : '',
797
+ branch && into ? `${branch}→${into}` : '',
798
+ validationStatus ? `validation=${validationStatus}` : '',
799
+ patchStatus ? `patch_equivalence=${patchStatus}` : '',
800
+ mergeStatus ? `merge=${mergeStatus}` : '',
801
+ convergenceStatus ? `final_convergence=${convergenceStatus}` : '',
802
+ ].filter(Boolean).join('; ');
803
+ return `[System] Refinery async job for ${args.nodeLabel} completed successfully${details ? ` (${details})` : ''}.\nNext step: ${nextStep}`;
804
+ }
805
+ if (args.event === 'refine:failed') {
806
+ const jobId = readRefineJobId({ metadataEvent: args.metadataEvent });
807
+ const result = readRecord(args.metadataEvent.result);
808
+ const validationSummary = readRecord(result?.validationSummary);
809
+ const patchEquivalence = readRecord(result?.patchEquivalence);
810
+ const finalConvergence = readRecord(result?.finalBranchConvergenceState);
811
+ const code = readNonEmptyString(result?.code);
812
+ const error = readNonEmptyString(result?.error);
813
+ const validationStatus = readNonEmptyString(validationSummary?.status);
814
+ const patchStatus = readNonEmptyString(patchEquivalence?.status)
815
+ || (patchEquivalence?.equivalent === true ? 'passed' : '');
816
+ const mergeStatus = result?.merged === true
817
+ ? 'merged'
818
+ : finalConvergence?.merged === false
819
+ ? 'not_merged'
820
+ : '';
821
+ const convergenceStatus = readNonEmptyString(result?.convergenceStatus)
822
+ || readNonEmptyString(finalConvergence?.status);
823
+ const blockedReason = readNonEmptyString(result?.blockedReason);
824
+ const nextStep = readNonEmptyString(result?.nextStep) || readNonEmptyString(finalConvergence?.nextStep);
825
+ const details = [
826
+ jobId ? `job_id=${jobId}` : '',
827
+ code ? `code=${code}` : '',
828
+ validationStatus ? `validation=${validationStatus}` : '',
829
+ patchStatus ? `patch_equivalence=${patchStatus}` : '',
830
+ mergeStatus ? `merge=${mergeStatus}` : '',
831
+ convergenceStatus ? `convergence=${convergenceStatus}` : '',
832
+ blockedReason ? `reason=${blockedReason}` : '',
833
+ ].filter(Boolean).join('; ');
834
+ const parts = [
835
+ `[System] Refinery async job for ${args.nodeLabel} failed${details ? ` (${details})` : ''}${error ? `: ${error}` : '.'}`,
836
+ nextStep ? `Next step: ${nextStep}` : 'Review the terminal refine event/ledger before retrying.',
837
+ ];
838
+ return parts.join('\n');
839
+ }
568
840
  return '';
569
841
  }
570
842
 
@@ -593,6 +865,28 @@ function injectMeshSystemMessage(components: DaemonComponents, args: {
593
865
  return { success: true, forwarded: 0, suppressed: true, intentionalCleanupStop: true };
594
866
  }
595
867
 
868
+ if (isDuplicateRefineTerminalEvent(args.meshId, args.event, args.metadataEvent)) {
869
+ LOG.info('MeshEvents', `Suppressed duplicate ${args.event} for refine job ${readRefineJobId({ metadataEvent: args.metadataEvent })}`);
870
+ return { success: true, forwarded: 0, suppressed: true, duplicateRefineTerminalEvent: true };
871
+ }
872
+
873
+ const eventTimestamp = readEventTimestamp(args.metadataEvent.timestamp);
874
+ if (args.event === 'agent:generating_completed' && eventSessionId) {
875
+ const duplicateCompletion = isDuplicateMeshCompletionEvent({
876
+ meshId: args.meshId,
877
+ event: args.event,
878
+ sessionId: eventSessionId,
879
+ providerType: readNonEmptyString(args.metadataEvent.providerType) || undefined,
880
+ providerSessionId: readNonEmptyString(args.metadataEvent.providerSessionId) || undefined,
881
+ timestamp: eventTimestamp,
882
+ finalSummary: readNonEmptyString(args.metadataEvent.finalSummary) || undefined,
883
+ });
884
+ if (duplicateCompletion) {
885
+ LOG.info('MeshEvents', `Suppressed duplicate completion for mesh ${args.meshId} session ${eventSessionId}`);
886
+ return { success: true, forwarded: 0, suppressed: true, duplicateCompletion: true };
887
+ }
888
+ }
889
+
596
890
  // ── Task Queue & Ledger ──
597
891
  let completedTaskForLedger: { id?: string } | null = null;
598
892
  if (args.event === 'agent:generating_completed') {
@@ -601,20 +895,27 @@ function injectMeshSystemMessage(components: DaemonComponents, args: {
601
895
  const providerType = readNonEmptyString(args.metadataEvent.providerType);
602
896
 
603
897
  if (sessionId) {
604
- const completedTask = updateSessionTaskStatus(args.meshId, sessionId, 'completed');
898
+ const completedTask = updateSessionTaskStatus(args.meshId, sessionId, 'completed', {
899
+ occurredAt: eventTimestamp !== null ? new Date(eventTimestamp).toISOString() : undefined,
900
+ });
605
901
  completedTaskForLedger = completedTask ? { id: completedTask.id } : null;
606
902
  if (nodeId && providerType) {
607
- // Short delay to allow completion event to propagate before pulling next
608
- setTimeout(() => {
903
+ // Queue state is already updated above; setImmediate avoids the
904
+ // 500 ms artificial delay while still deferring past this call frame.
905
+ setImmediate(() => {
609
906
  tryAssignQueueTask(components, args.meshId, nodeId, sessionId, providerType);
610
- }, 500);
907
+ });
611
908
  }
612
909
  }
613
910
  } else if (args.event === 'agent:ready') {
614
911
  const sessionId = resolveEventSessionId(args.metadataEvent, args.sourceInstanceId);
615
912
  const nodeId = readNonEmptyString(args.nodeId) || readNonEmptyString(args.metadataEvent.meshNodeId);
616
913
  const providerType = readNonEmptyString(args.metadataEvent.providerType);
617
- const completedTask = sessionId
914
+ const providerSessionId = readNonEmptyString(args.metadataEvent.providerSessionId) || undefined;
915
+ const finalSummary = readNonEmptyString(args.metadataEvent.finalSummary) || undefined;
916
+ const workerResult = readWorkerResultMetadata(args.metadataEvent);
917
+ const hasCompletionEvidence = !!finalSummary || !!workerResult;
918
+ const completedTask = sessionId && hasCompletionEvidence
618
919
  ? updateSessionTaskStatus(args.meshId, sessionId, 'completed')
619
920
  : null;
620
921
  if (completedTask) {
@@ -630,15 +931,17 @@ function injectMeshSystemMessage(components: DaemonComponents, args: {
630
931
  nodeLabel: args.nodeLabel,
631
932
  taskId: completedTask.id,
632
933
  completedViaReady: true,
633
- providerSessionId: readNonEmptyString(args.metadataEvent.providerSessionId) || undefined,
634
- finalSummary: readNonEmptyString(args.metadataEvent.finalSummary) || undefined,
934
+ providerSessionId,
935
+ finalSummary,
936
+ workerResult,
635
937
  evidence: buildTaskCompletionEvidence({
636
938
  event: 'agent:ready',
637
939
  nodeId,
638
940
  sessionId,
639
941
  providerType: providerType || undefined,
640
- providerSessionId: readNonEmptyString(args.metadataEvent.providerSessionId) || undefined,
641
- finalSummary: readNonEmptyString(args.metadataEvent.finalSummary) || undefined,
942
+ providerSessionId,
943
+ finalSummary,
944
+ workerResult,
642
945
  }),
643
946
  },
644
947
  });
@@ -648,13 +951,15 @@ function injectMeshSystemMessage(components: DaemonComponents, args: {
648
951
  }
649
952
 
650
953
  if (sessionId && nodeId && providerType) {
651
- remoteIdleSessions.set(`${nodeId}:${sessionId}`, { nodeId, sessionId, providerType });
652
- setTimeout(() => {
954
+ sweepExpiredRemoteIdleSessions();
955
+ remoteIdleSessions.set(`${nodeId}:${sessionId}`, {
956
+ nodeId, sessionId, providerType,
957
+ expiresAt: Date.now() + REMOTE_IDLE_SESSION_TTL_MS,
958
+ });
959
+ setImmediate(() => {
653
960
  const assigned = tryAssignQueueTask(components, args.meshId, nodeId, sessionId, providerType);
654
- if (assigned) {
655
- remoteIdleSessions.delete(`${nodeId}:${sessionId}`);
656
- }
657
- }, 500);
961
+ if (assigned) remoteIdleSessions.delete(`${nodeId}:${sessionId}`);
962
+ });
658
963
  }
659
964
  } else if (args.event === 'agent:generating_started') {
660
965
  const sessionId = resolveEventSessionId(args.metadataEvent, args.sourceInstanceId);
@@ -669,7 +974,8 @@ function injectMeshSystemMessage(components: DaemonComponents, args: {
669
974
  remoteIdleSessions.delete(`${nodeId}:${sessionId}`);
670
975
  }
671
976
  if (sessionId) {
672
- updateSessionTaskStatus(args.meshId, sessionId, 'failed');
977
+ const failedTask = updateSessionTaskStatus(args.meshId, sessionId, 'failed');
978
+ completedTaskForLedger = failedTask ? { id: failedTask.id } : null;
673
979
  }
674
980
  }
675
981
 
@@ -679,14 +985,18 @@ function injectMeshSystemMessage(components: DaemonComponents, args: {
679
985
  const ledgerNodeId = readNonEmptyString(args.nodeId) || readNonEmptyString(args.metadataEvent.meshNodeId) || undefined;
680
986
  const ledgerSessionId = resolveEventSessionId(args.metadataEvent, args.sourceInstanceId) || undefined;
681
987
  const ledgerProviderType = readNonEmptyString(args.metadataEvent.providerType) || undefined;
988
+ const providerSessionId = readNonEmptyString(args.metadataEvent.providerSessionId) || undefined;
989
+ const finalSummary = readNonEmptyString(args.metadataEvent.finalSummary) || undefined;
990
+ const workerResult = readWorkerResultMetadata(args.metadataEvent);
682
991
  const completionEvidence = ledgerKind === 'task_completed' && ledgerNodeId && ledgerSessionId
683
992
  ? buildTaskCompletionEvidence({
684
993
  event: 'agent:generating_completed',
685
994
  nodeId: ledgerNodeId,
686
995
  sessionId: ledgerSessionId,
687
996
  providerType: ledgerProviderType,
688
- providerSessionId: readNonEmptyString(args.metadataEvent.providerSessionId) || undefined,
689
- finalSummary: readNonEmptyString(args.metadataEvent.finalSummary) || undefined,
997
+ providerSessionId,
998
+ finalSummary,
999
+ workerResult,
690
1000
  })
691
1001
  : undefined;
692
1002
  appendLedgerEntry(args.meshId, {
@@ -698,8 +1008,12 @@ function injectMeshSystemMessage(components: DaemonComponents, args: {
698
1008
  event: args.event,
699
1009
  nodeLabel: args.nodeLabel,
700
1010
  taskId: completedTaskForLedger?.id || undefined,
701
- providerSessionId: readNonEmptyString(args.metadataEvent.providerSessionId) || undefined,
702
- finalSummary: readNonEmptyString(args.metadataEvent.finalSummary) || undefined,
1011
+ providerSessionId,
1012
+ finalSummary,
1013
+ workerResult,
1014
+ completionDiagnostic: args.metadataEvent.completionDiagnostic && typeof args.metadataEvent.completionDiagnostic === 'object'
1015
+ ? args.metadataEvent.completionDiagnostic
1016
+ : undefined,
703
1017
  evidence: completionEvidence,
704
1018
  },
705
1019
  });
@@ -772,6 +1086,14 @@ function injectMeshSystemMessage(components: DaemonComponents, args: {
772
1086
  }
773
1087
  }
774
1088
 
1089
+ const messageText = buildMeshSystemMessage({
1090
+ event: args.event,
1091
+ nodeLabel: args.nodeLabel,
1092
+ metadataEvent: args.metadataEvent,
1093
+ recoveryContext,
1094
+ });
1095
+ if (!messageText) return { success: false, error: 'unsupported mesh event' };
1096
+
775
1097
  const coordinatorInstances = components.instanceManager.getByCategory('cli').filter((inst) => {
776
1098
  const instState = inst.getState();
777
1099
  if (instState.settings?.meshCoordinatorFor !== args.meshId) return false;
@@ -779,31 +1101,53 @@ function injectMeshSystemMessage(components: DaemonComponents, args: {
779
1101
  return true;
780
1102
  });
781
1103
 
1104
+ // Refine terminal events (refine:completed, refine:failed) are coordinator-delivered
1105
+ // synchronously; only buffer them for MCP when no CLI coordinator is present.
1106
+ // Agent runtime events (agent:*) use dual delivery so both CLI and MCP coordinators
1107
+ // receive them regardless of whether a live CLI coordinator session is active.
1108
+ const isRefineTerminalEvent = REFINE_TERMINAL_EVENTS.has(args.event);
1109
+
782
1110
  if (coordinatorInstances.length === 0) {
783
1111
  // No CLI coordinator session found — buffer for MCP-based coordinators.
784
- if (pendingMeshCoordinatorEvents.length < MAX_PENDING_EVENTS) {
785
- pendingMeshCoordinatorEvents.push({
1112
+ if (queuePendingMeshCoordinatorEvent({
786
1113
  event: args.event,
787
1114
  meshId: args.meshId,
788
1115
  nodeLabel: args.nodeLabel,
1116
+ nodeId: args.nodeId || undefined,
1117
+ workspace: readNonEmptyString(args.metadataEvent.workspace),
789
1118
  metadataEvent: {
790
1119
  ...args.metadataEvent,
791
1120
  ...(recoveryContext ? { recoveryContext } : {}),
792
1121
  },
1122
+ coordinatorMessage: messageText,
793
1123
  queuedAt: Date.now(),
794
- });
1124
+ })) {
795
1125
  LOG.info('MeshEvents', `Queued ${args.event} for MCP coordinator (mesh ${args.meshId})`);
796
1126
  }
797
1127
  return { success: true, forwarded: 0 };
798
1128
  }
799
1129
 
800
- const messageText = buildMeshSystemMessage({
801
- event: args.event,
802
- nodeLabel: args.nodeLabel,
803
- metadataEvent: args.metadataEvent,
804
- recoveryContext,
805
- });
806
- if (!messageText) return { success: false, error: 'unsupported mesh event' };
1130
+ // CLI coordinator is present. For non-refine events, also buffer for MCP coordinators
1131
+ // that poll via get_pending_mesh_events (dual delivery). Refine terminal events are
1132
+ // forwarded directly only — they must not accumulate in the pending queue when a live
1133
+ // coordinator already received them.
1134
+ if (!isRefineTerminalEvent) {
1135
+ if (queuePendingMeshCoordinatorEvent({
1136
+ event: args.event,
1137
+ meshId: args.meshId,
1138
+ nodeLabel: args.nodeLabel,
1139
+ nodeId: args.nodeId || undefined,
1140
+ workspace: readNonEmptyString(args.metadataEvent.workspace),
1141
+ metadataEvent: {
1142
+ ...args.metadataEvent,
1143
+ ...(recoveryContext ? { recoveryContext } : {}),
1144
+ },
1145
+ coordinatorMessage: messageText,
1146
+ queuedAt: Date.now(),
1147
+ })) {
1148
+ LOG.info('MeshEvents', `Queued ${args.event} for MCP coordinator (mesh ${args.meshId})`);
1149
+ }
1150
+ }
807
1151
 
808
1152
  for (const coord of coordinatorInstances) {
809
1153
  const coordState = coord.getState();
@@ -834,6 +1178,15 @@ export function handleMeshForwardEvent(components: DaemonComponents, payload: Re
834
1178
  providerType: readNonEmptyString(payload.providerType),
835
1179
  providerSessionId: readNonEmptyString(payload.providerSessionId),
836
1180
  finalSummary: readNonEmptyString(payload.finalSummary) || readNonEmptyString(payload.summary),
1181
+ jobId: readNonEmptyString(payload.jobId),
1182
+ interactionId: readNonEmptyString(payload.interactionId),
1183
+ status: readNonEmptyString(payload.status),
1184
+ targetDaemonId: readNonEmptyString(payload.targetDaemonId),
1185
+ startedAt: readNonEmptyString(payload.startedAt),
1186
+ completedAt: readNonEmptyString(payload.completedAt),
1187
+ retryOfJobId: readNonEmptyString(payload.retryOfJobId),
1188
+ ...(payload.result && typeof payload.result === 'object' && !Array.isArray(payload.result) ? { result: payload.result } : {}),
1189
+ ...(payload.timestamp !== undefined ? { timestamp: payload.timestamp } : {}),
837
1190
  intentional: payload.intentional === true,
838
1191
  intentionalStop: payload.intentionalStop === true,
839
1192
  operatorCleanup: payload.operatorCleanup === true,